Asymmetric multilevel outphasing architecture for multistandard transmitters
The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.
Citation
SungWon Chung et al. “Asymmetric multilevel outphasing
architecture for multi-standard transmitters.” Radio Frequency
Integrated Circuits Symposium, 2009. RFIC 2009. IEEE. 2009.
237-240. © 2009 Institute of Electrical and Electronics Engineers.
As Published
http://dx.doi.org/10.1109/RFIC.2009.5135530
Publisher
Institute of Electrical and Electronics Engineers
Version
Final published version
Accessed
Thu May 26 08:47:51 EDT 2016
Citable Link
http://hdl.handle.net/1721.1/54242
Terms of Use
Article is made available in accordance with the publisher's policy
and may be subject to US copyright law. Please refer to the
publisher's site for terms of use.
Detailed Terms
RMO3C-3
Asymmetric Multilevel Outphasing Architecture for
Multi-standard Transmitters
SungWon Chung, Philip A. Godoy, Taylor W. Barton, Everest W. Huang † , David J. Perreault, and Joel L. Dawson
Microsystems Technology Laboratories, Massachusetts Institute of Technology, Cambridge, MA 02139
†
Lincoln Laboratory, Massachusetts Institute of Technology, Lexington, MA 02420
Multilevel DC/DC
Abstract— We describe a new outphasing transmitter architecture in which the supply voltage for each PA can switch
among multiple levels. It is based on a new asymmetric
multilevel outphasing (AMO) modulation technique which
increases overall efficiency over a much wider output power
range than the standard LINC system while maintaining
high linearity. For demonstration, the overall transmitter
is simulated in a 65nm CMOS process with HSUPA and
WLAN signals. The simulation results show an efficiency
improvement from 17.7% to 40.7% for HSUPA at 25.3dBm
output power and from 11.3% to 35.5% for WLAN 802.11g
at 22.8dBm while still meeting system linearity requirements.
A
φ
φ1
φ2
Digital−to−RF
Phase Converter
(DRFPC)
PA1
Combiner
PA2
Predistorter
Training
Fig. 1.
Asymmetric multilevel outphasing (AMO) architecture.
expect significant improvement over conventional techniques, with a resulting system that is compatible with
a wide range of communications standards.
In this paper we first present a summary of conventional
approaches to power transmission in Section II. In Section
III we introduce the AMO modulation technique, and in
Section IV we describe the AMO transmitter architecture.
Promising system-level simulation results are presented
and discussed in Section V, with conclusions following
in Section VI.
Index Terms— outphasing, LINC, wideband phase modulator, digital predistortion.
I. I NTRODUCTION
The primary challenge in RF transmitter design is
centered around a design tradeoff between the linearity of
the power amplifier (PA) and its efficiency. This tradeoff
relates directly to the usefulness of the resulting device:
high linearity results in a higher possible data rate and
therefore compatibility with complex standards such as
WLAN/WiMAX, and high efficiency allows for either
longer use or smaller battery size (e.g., in cell phone
applications). The general perception that the tradeoff
between linearity and efficiency is fundamental tends to
produce designs that compromise between the two ideals.
The resulting systems may be either linear or efficient,
or are designed specifically for a single communications
standard and therefore have limited flexibility of use.
Meanwhile, consumer demand for both greater transmission rates and smaller devices continues to drive the need
for an architecture that is capable of both linearity and
efficiency.
We propose the Asymmetric Multilevel Outphasing
(AMO) transmitter architecture, described in this paper,
as a solution to this linearity-efficiency tradeoff problem.
This architecture, shown in Fig. 1, results in significant
efficiency improvement over previous methods. Our approach is based on several innovations, including using
asymmetric power supplies in a LINC (linear amplification
using nonlinear components)-like configuration and an
all-digital AMO modulator. Using these techniques, we
II. C ONVENTIONAL A PPROACHES
Communication standards that support high data rates
such as WLAN/WiMAX employ variable-envelope modulation, and so linear amplification is required. One approach is to use an inefficient but highly linear PA.
However, there are two main types of transmitter architectures that enable the use of more efficient but nonlinear
switching-mode PAs: (1) polar, and (2) outphasing, or
LINC.
The fundamental idea of polar architectures, shown in
Fig. 2(a), is to divide the signal to be amplified into
amplitude and phase components. The phase component is
used as the input to a nonlinear, high-efficiency switching
PA, while the amplitude component drives the power
supply of the PA to create a varying-envelope signal. While
this improves the PA efficiency, it also requires the use
of an efficient power converter. Because power converter
efficiency degrades dramatically as bandwidth increases, it
is very difficult to achieve high efficiency for high data-rate
communication standards. This is exacerbated by the 5-10x
bandwidth expansion that occurs during the conversion
from Cartesian to polar coordinates [1]. Thus this method
is only practical for low-bandwidth systems.
Outphasing [2], and specifically the LINC architecture,
introduced by Cox in [3], is shown in Fig. 2(b). It is based
This work was sponsored by the Department of the Air Force under
Contract FA8271-05-C-0002. Opinions, interpretations, conclusions, and
recommendations are those of the author and are not necessarily endorsed
by the United States Government. This work was also funded in part by
the MIT Deshpande Center.
978-1-4244-3376-6/978-1-4244-3378-0/09/$25.00 © 2009 IEEE
AMO Modulator
with
Predistortion
Switch
Switch
A1
A2
237
2009 IEEE Radio Frequency Integrated Circuits Symposium
Authorized licensed use limited to: MIT Libraries. Downloaded on April 05,2010 at 14:09:39 EDT from IEEE Xplore. Restrictions apply.
amplitude
g replacements
PA
polar
conv.
PA
Conventional multi-standard transmitters: (a) Polar (b) LINC.
Q
LINC
I
Q
ML−LINC
Q
0.03
0.02
40
0.01
20
0
−20
AMO
I
−15
−10
−5
Normalized Output Power (dB)
0
0
Fig. 4. Amplitude distribution of HSUPA signal and corresponding
efficiency for standard LINC, ML-LINC, and AMO.
I
0.04
Probability Density
Fig. 3. Signal component vector diagram for LINC, ML-LINC, and
AMO. The smallest outphasing angle is achieved with AMO.
on the idea that an arbitrary input signal can be divided
into two constant-amplitude, phase-modulated signals that
can each be non-linearly amplified and then passively
recombined as a vector sum to produce an output signal
that is a linearly amplified version of the input. The LINC
strategy eliminates the high-bandwidth power converter of
the basic polar architecture, using outphasing to realize
amplitude variation. However, the efficiency of the power
combining is high only over a small range of output
powers. To avoid signal distortion and preserve switching amplifier efficiency, an isolating combiner such as a
Wilkinson combiner must be used. Isolating combiners
achieve 100% efficiency only at maximum output power.
When the inputs are outphased to vary the amplitude,
power is wasted as heat in the isolation resistor [4].
The overall efficiency is therefore inversely proportional
to the peak-to-average power ratio (PAPR), limiting the
benefits of this technique in high data-rate communication
standards such as WiMAX, in which the PAPR is high.
80
LINC
ML−LINC
AMO
PAE
PDF
0.03
60
0.02
40
0.01
20
0
−20
−15
−10
−5
Normalized Output Power (dB)
0
0
Fig. 5. Amplitude distribution of WLAN signal and corresponding
efficiency for standard LINC, ML-LINC, and AMO.
In order to have linear PA output, C(t) is predistorted into
P (t) = Ap (t) ejθp (t)
(2)
by a polar lookup table (LUT). P (t) is decomposed into
two parts as
P (t) = W V1 (t) ejφ1 (t) , V2 (t) ejφ2 (t)
(3)
where W represents Wilkinson power combining and
III. P ROPOSED A PPROACH
A. AMO Modulation
Fundamentally, AMO modulation decomposes a complex vector, which represents a baseband constellation
point, into two vectors such that the sum of the two vectors
constructs the original complex vector with the minimum
outphasing angle, as illustrated in Fig. 3. The two vectors
are the baseband representation of the two PA outputs. As
compared to the multilevel LINC (ML-LINC) technique
[5], by making independent changes in the supply voltage
for each of the two PAs, the AMO technique results in
smaller outphasing angles so that higher efficiency can be
achieved even in relatively high-PAPR standards.
Mathematically, AMO modulation can be defined with
a polar representation of a baseband signal,
C(t) = ri (t) + jrq (t) = A(t) ejθ(t) .
60
Total Efficiency (%)
Σ
(b)
(a)
Fig. 2.
∆
80
LINC
ML−LINC
AMO
PAE
PDF
Total Efficiency (%)
Riso
S2 (t)
0.04
PA
Probability Density
Q
Sin (t)
A(t) amplitude
control
polar
conv. φ(t) phase
PA
control
signal separation
I
S1 (t)
−1
φ1 (t)
= θp (t) + cos
φ2 (t)
= θp (t) − cos−1
V1 (t)2 + 4Ap (t)2 − V2 (t)2
,
4V1 (t)Ap (t)
V2 (t)2 + 4Ap (t)2 − V1 (t)2
.
4V2 (t)Ap (t)
B. Multi-standard Efficiency Optimization
As we use AMO to minimize loss in the Wilkinson
combiner, we have to determine the optimal value of each
level rk (these are the maximum output amplitudes for
each of the different supply voltage levels when both PAs
are driven by the same supply). The Wilkinson combiner
efficiency at a given output amplitude A and driven by two
PAs with different supply voltages is given by

2  r +r 2 
2 k2 j 
A



(4)
ηc (A, rk , rj ) = 
 2
rk +rj
rk + rj2
(1)
2
238
Authorized licensed use limited to: MIT Libraries. Downloaded on April 05,2010 at 14:09:39 EDT from IEEE Xplore. Restrictions apply.
Predistortion
for DRFPC/Switch/PA Linearization
AMO Modulator
Digital Back−end
AMO Modulator
Analog/RF Front−end
Amplitude Switch & RF PA
VDD
AM/Lvl
LUT
Q
COR−
DIC
A
φ
Interval
Peak
Detect
Polar LUT
Predistortion
Fig. 6.
AMO Mapper
I
+
rk +rk+1
2
Combiner
PA1
DRFPC
DRFPC
PA2
Φ1(t)
PA
0
0
180
0
Α2(t)
PA
Φ2(t)
RISO
Fig. 7.
Amplitude switching network and PAs.
from DRFPC (digital-to-RF phase converter)/Switch/PA,
2) AMO modulator, and 3) amplitude switch and RF PA.
The AMO modulator first determines the combination
of the two PA supply voltages based on a peak amplitude
within a time interval. The AMO mapper decomposes
the predistorted amplitude and phase into two pairs of
amplitude and phase commands. The AMO mapper is
implemented with a first order approximation of (3). The
time delay mismatch between the amplitude and phase
paths is maintained to within less than 1ns by a time
aligner between the AMO mapper and the amplitude
switch.
The DRFPC, which is based on the direct-digital RF
modulator [6], performs phase modulation by embedding
the phase component of the AMO mapper output into an
RF carrier. The DRFPC consists of an array of current
steering switches as in the direct-digital RF modulator. The
DRFPC brings a significant transmitter power efficiency
boost particularly for low output power levels for two
reasons. First, the analog matching requirement in the
current steering switches is relaxed because the static
phase errors in the DRFPC output, which result from
analog mismatch, can be corrected by the predistorter.
Second, as compared to traditional IQ modulators, the
DRFPC does not need baseband active filters for DAC
output shaping.
Fig. 7 shows the schematic of the PA supply modulator,
which consists of a fast switching network driven by the
AMO modulator. For maximum efficiency, the PAs are
switching-mode amplifiers (e.g., class-E, class-F, class-
p(A) ηc (A, rk , rk+1 ) ηP A (rk , rk+1 ) dA
p(A) ηc (A, rk+1 , rk+1 ) ηP A (rk+1 , rk+1 ) dA
Antenna
Α1(t)
rk +rk+1
2
rk
k=1
Z rk+1
4−to−1
Switch
Switch
Supply Voltages
0
+
φ1
φ2
AM/PM
Time Align
Time Align
AMO transmitter block diagram.
Note that this simplifies to the standard Wilkinson efficiency when rk = rj . The total average efficiency can
be computed if the amplitude PDF p(A) of the signal is
known (see Fig. 4 and Fig. 5 for the PDFs of HSUPA and
WLAN signals, respectively). This is done by dividing
the PDF into several regions separated by the rk (and
their combinations), integrating the PDF curve to find the
efficiency in each region, and summingthe result. For N
different supply voltages, there are N2 combinations of
supply voltages for the two PAs. However, the combiner
efficiency decreases as the difference between two levels
increases. Also, the efficiency improvement is small when
the difference between the two levels is large. Therefore,
in our system we restrict the combinations to be adjacent
supply levels (i.e., rk and rk+1 ). With this restriction, the
total average efficiency can be computed as
Z r1
ηtot =
p(A) ηc (A, r1 , r1 ) ηP A (r1 , r1 ) dA
N
−1 h Z
X
A1
A2
i
where ηP A (rk , rj ) is the total PA efficiency at the two
supply voltages rk and rj . Using this equation, the optimum set of supply voltage levels for a given amplitude
PDF can be found by exhaustive search.
Fig. 4 shows the amplitude PDF for a HSUPA signal and the corresponding optimum efficiency curves for
standard LINC, ML-LINC, and AMO using 4 different
supply levels.1 The figure shows that ML-LINC increases
the efficiency over a much wider power range than the
standard LINC system, and AMO increases this range even
further by effectively doubling the number of levels. Fig. 5
shows the amplitude PDF and optimum efficiency curves
for the WLAN signal.
IV. AMO A RCHITECTURE
Fig. 6 shows the AMO architecture, which consists of
1) predistorter for linearizing the combined nonlinearity
1 The PA efficiency curve was obtained from simulation, which will
be described in Section V.
239
Authorized licensed use limited to: MIT Libraries. Downloaded on April 05,2010 at 14:09:39 EDT from IEEE Xplore. Restrictions apply.
22.8−dBm WLAN 802.11g OFDM
−10
−20
Transmitter
spectral mask
AMO
−30 AMO w/ DPD
−40
−50
1790
1810
1830 1850 1870
Frequency (MHz)
1890
−20
AMO
−30
−40 AMO w/ DPD
−60
1910
Fig. 8. Simulated spectrum for WLAN 802.11g transmission with 22.8dBm channel power, 20-MHz channel bandwidth, and 7.0-dB PAPR.
AMO EVM = −27.85dB
Normalized Q−channel
Normalized Q−channel
−10
−50
−60
1
25.3−dBm UMTS HSUPA
0
Spectrum (dBr)
Spectrum (dBr)
0
0.5
0
−0.5
−1
−1 −0.5
0
0.5
1
Normalized I−channel
1
1835
1840
1845 1850 1855
Frequency (MHz)
1860
1865
Fig. 10. Simulated spectrum for HSUPA transmission with 25.3-dBm
channel power, 5-MHz channel bandwidth, and 5.0-dB PAPR.
TABLE I
P ERFORMANCE C OMPARISON FOR LINC, ML-LINC, AND
Wireless
Linearity
Architecture
EVM
Standard
Margin
LINC
-38.5dB
7.9dB
ML-LINC
-33.6dB
0.1dB
HSUPA
ML-LINC w/ DPD
-42.4dB
7.2dB
(PAPR 5dB)
AMO
-33.9dB
0.2dB
AMO w/ DPD
-42.3dB
9.0dB
LINC
-34.7dB
9.0dB
WLAN
ML-LINC
-26.1dB
-4.7dB
802.11g
ML-LINC w/ DPD
-35.6dB
4.8dB
(PAPR 7dB)
AMO
-27.8dB
-2.7dB
AMO w/ DPD
-36.3dB
3.3dB
AMO w/ DPD EVM = −36.25dB
0.5
0
−0.5
−1
−1 −0.5
0
0.5
1
Normalized I−channel
Fig. 9. EVM for the WLAN 802.11g 54Mbps OFDM transmission
(BPSK constellations for frame preambles are shown in red).
AMO
ηavg
17.7%
37.7%
37.1%
41.1%
40.7%
11.3%
31.9%
30.9%
37.1%
35.5%
E/F, etc.).
V. S IMULATION R ESULTS
maintain high linearity by using digital predistortion. The
AMO modulation technique was described, along with
the optimization procedure for the supply voltages based
on the envelope distribution of the modulated signal.
Finally, we demonstrate in simulation a four-level AMO
transmitter as a design example which improves the overall
efficiency from 17.7% to 40.7% for HSUPA and from
11.3% to 35.5% for WLAN 802.11g while still meeting
system linearity requirements.
To demonstrate the feasibility of the AMO transmitter,
the system was simulated using Cadence Spectre and
Agilent ADS using 4 different supply voltages. The PAs
and switching network were designed in a 65nm CMOS
process to model the main sources of power loss. The
fully differential PAs were implemented as cascoded classE amplifiers with inverters for the pre-drivers and adaptive
biasing for the cascodes to increase efficiency [7]. Both the
PA cascode transistors and the multi-level switching network were implemented with thick oxide devices. All other
blocks in the system were implemented using Verilog-A.
For WLAN, digital predistortion (DPD) linearizes both the
transmit spectrum (Fig. 8) and EVM (Fig. 9) by more than
8 dB. Fig. 10 shows more than 10-dB improvement in the
HSUPA transmit spectrum after predistortion is applied.
Table I summarizes the EVM, ACLR (adjacent channel
leakage ratio) margin for HSUPA, transmit spectral mask
margin for WLAN, and average power efficiency of the
proposed AMO transmitter in comparison with LINC and
ML-LINC. The efficiency considers 1-dB insertion loss
from the Wilkinson combiner as well as the power loss
from the switch and the PA.
R EFERENCES
[1] F. Wang et al., “Wideband envelope elimination and restoration
power amplifier with high efficiency wideband envelope amplifier
for WLAN 802.11g applications,” in Proc. IEEE Int’l Microwave
Symp., 2005, pp. 645–648.
[2] H. Chireix, “High-power outphasing modulation,” Proc. IRE, vol. 23,
pp. 1370–1392, 1935.
[3] D. Cox, “Linear amplification with nonlinear components,” IEEE
Trans. on Communications, pp. 1942–1945, Dec. 1974.
[4] I. Hakala et al., “A 2.14-GHz Chireix outphasing transmitter,” IEEE
Trans. Microwave Theory Tech., pp. 2129–2138, June 2005.
[5] Y.-J. Chen et al., “Multilevel LINC system design for wireless
transmitters,” in Int’l Symp. on VLSI-DAT, 2007.
[6] P. Eloranta et al., “A multimode transmitter in 0.13um CMOS using
direct-digital RF modulator,” IEEE J. Solid-State Circuits, pp. 2774–
2784, Dec. 2007.
[7] M. Apostolidou et al., “A 65nm CMOS 30dBm class-E power
amplifier with 60% power added efficiency,” in Proc. IEEE RFIC
Symp., 2008, pp. 141–144.
VI. C ONCLUSIONS
The AMO transmitter architecture was proposed to
not only greatly increase transmitter efficiency but also
240
Authorized licensed use limited to: MIT Libraries. Downloaded on April 05,2010 at 14:09:39 EDT from IEEE Xplore. Restrictions apply.